Deserialization Threats
PAIT-ARV-100 (Archive Slip Bugs in ML Models)
Ecenur ÜzeJunior Pentester
September 2, 2024
1 min read
Overview
Many machine learning models are essentially zipped archives containing weights and configurations. A malicious archive can include filenames like ../../../../etc/passwd. When a Python script uses a vulnerable unarchiving function to extract the model, it overwrites critical files on the host.
Remediation
Sanitize paths during extraction. Never extract archives directly to root. Use safe unzipping libraries.
Security Validation
Have you tested this risk in your own system?
Eresus Security delivers real exploit evidence through penetration testing, AI agent security, and red team operations.
Request a pilot test